JP2000222406A - Voice recognition and translation device and its method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the accuracy of processing by using scores in voice recognition processing and translation processing and to speed up the voice recognition and translation processing by reducing unnecessary processing. SOLUTION: The voice recognition and translation device is provided with a voice recognition means for executing voice recognition processing by applying an acoustic model and a recognition language model to an inputted voice, a voice recognition score generation means for generating a voice recognition score indicating the processing accuracy of voice recognition processing based on a voice recognition result and a translation means for executing translation processing for translating the language of the inputted voice to a different language by using the voice recognition result. The device is also provided with a translation score generation means for generating a translation score indicating the processing accuracy of translation processing, a synthetic score generation means for generating a synthetic score indicating the processing accuracy of voice recognition processing and translation processing by using the voice recognition score outputted from the voice recognition score generation means and the translation score outputted from the translation score generation means and an evaluation means for evaluating the translation result outputted from the translation means.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition translation apparatus and method for translating a natural language input by speech into another natural language.

[0002]

2. Description of the Related Art In general, a speech recognition and translation apparatus that translates a natural language (source language) input by speech into another natural language (target language) and outputs the speech and character strings, etc. The two-stage processing is combined. This speech recognition and translation device recognizes a speech signal for each word by performing acoustic analysis by referring to an acoustic model and analyzing a speech signal in which an input language is detected.
Then, the speech recognition and translation device performs a process of collating each word using word dictionary information that describes the relationship between the speech signal and semantic information such as the part of speech of the word, thereby forming a word string formed by connecting a plurality of words. Generate candidates. At this time, the speech recognition translation apparatus adds a speech recognition score indicating the linguistic appearance probability of the word string candidate generated based on the degree of coincidence with the acoustic model.

[0003] However, in the above-described speech recognition and translation apparatus, it is difficult to achieve high recognition accuracy with a small number of candidates by performing speech recognition processing using an acoustic model. Therefore, in the speech recognition translation apparatus, in the speech recognition processing, correction of the speech recognition score for words (or word strings) using linguistic information and narrowing of candidates are often performed.

[0004] A recognition language model indicating the connection relation of each word used in the speech recognition processing is a statistical language model, particularly a word model of a word created from corpus information indicating a large number of sample sentences, because of easy integration with an acoustic model. Chain probability is often used. As for the processing form, the efficiency of the processing is improved by treating the score calculation in the word string search processing using the acoustic model including the score of the statistical language model.

On the other hand, the translation process performs a translation process using a word string candidate generated by the speech recognition process. The speech recognition translation apparatus performs a translation process by analyzing a source language based on abstract language rules and performing a translation process, and a large amount of bilingual corpus from a source language to a target language. There is an example-based translation processing method for collecting translations and performing translation processing based thereon.

The above-described rule-based translation processing method has an advantage that a large coverage can be obtained by describing a small number of language rules. On the other hand, the example-based translation processing method has an advantage that an appropriate translation result depending on an individual example can be obtained.

In this speech recognition and translation apparatus, a score is added to the translation result obtained by translating the input speech on a linguistic scale in the translation processing, and the translation result is selected based on the score added to the translation result. Prioritization is done.

[0008]

In the above-described speech recognition and translation apparatus, the score in the speech recognition process and the score in the translation process are separately generated, and the accuracy of the process is evaluated by evaluating the processing accuracy in each process. Was improved. That is, in the above-described speech recognition and translation device, when evaluating the translation result obtained by the translation process obtained for the input speech, the score in the speech recognition process is often not reflected in the score of the final translation result. Therefore, in the above-described speech recognition translator,
It has been difficult to improve the accuracy of the final translation result using the correlation between the score of the speech recognition process and the score of the translation process.

Further, in the speech recognition and translation device, when performing control such as stopping the analysis process for a word string candidate that is unlikely to be a correct answer during the translation process, the analysis process is suspended including the score of the speech recognition process. In many cases, such decisions have not been made. Judgment based solely on such linguistic scoring has to relax the criteria for stopping the processing in order to reduce the risk of stopping the analysis of a word string candidate that is originally a correct answer. There is a waste of processing of generating a translation result by translation processing up to the candidate. Therefore, in such a speech recognition and translation device, there has been a problem that the processing amount in the translation process cannot be reduced.

Accordingly, the present invention has been proposed in view of the above-described circumstances, and improves the processing accuracy by using the scores in the speech recognition processing and the translation processing, and reduces unnecessary processing. An object of the present invention is to provide a speech recognition translation apparatus and method capable of realizing high-speed speech recognition translation processing.

[0011]

According to the present invention, there is provided a speech recognition translating apparatus and method for solving the above-mentioned problem, wherein an acoustic model referred to for recognizing a waveform shape of a speech signal as speech for an inputted speech. And performing a speech recognition applying a recognition language model indicating a connection relationship between words constituting the sentence, generating a speech recognition score indicating a processing accuracy of the speech recognition for the speech recognition result, and inputting the speech using the speech recognition result. Translated into a language different from the input speech, generate a translation score indicating the translation processing accuracy of the translation result, and indicate the processing accuracy of the speech recognition process and the translation process using the speech recognition score and the translation score. It is characterized in that a combined score is generated, and a translation result to be output is evaluated using the combined score.

According to such a speech recognition translation apparatus and method, a translation result to be output is evaluated using a synthesized score generated using the speech recognition score and the translation score.

[0013]

Embodiments of the present invention will be described below in detail with reference to the drawings.

The present invention is applied to, for example, a speech recognition and translation device 1 configured as shown in FIG. The speech recognition and translation device 1 includes a speech recognition unit 2 to which external speech to be subjected to speech recognition processing and translation processing is input. The voice recognition unit 2 detects a voice signal obtained by detecting a voice generated by an acoustoelectric conversion mechanism (not shown) including, for example, a microphone for detecting a voice. The speech recognition unit 2 performs a speech recognition process on the input speech signal using the acoustic model information stored in the acoustic model storage unit 3 and the recognition language model information stored in the recognition language model storage unit 4.

The acoustic model storage unit 3 includes a speech recognition unit 2
By reading the contents, the stored acoustic model information is output to the speech recognition unit 2. The acoustic model information stored in the acoustic model storage unit 3 is information referred to for recognizing a waveform shape of an audio signal as audio. Specifically, as shown in FIG. 2, the acoustic model information is information for recognizing a word “I” when a voice “watashi” is input, for example.

The acoustic model storage unit 3 stores word dictionary information indicating semantic information of each word constituting a sentence. The word dictionary information is, for example, information indicating the part of speech of each word as semantic information, and is information for recognizing the word "I" as a "pronoun" as shown in FIG.

When the contents of the recognition language model storage unit 4 are read by the speech recognition unit 2, the stored recognition language model information is output to the speech recognition unit 2.
The recognition language model storage unit 4 stores recognition language model information describing the connection relation of each word. As the recognition language model information, for example, as shown in FIG.
Information for connecting three words (three chains of words) is stored, and a score based on the appearance probability of each word is added to each piece of information.

The speech recognition unit 2 is configured to determine the acoustic model information stored in the acoustic model storage unit 3 and the recognition language stored in the recognition language model storage unit 4 with respect to the speech signal of the utterance of the original language by the input speech. The speech recognition processing is performed by referring to information necessary for speech recognition of the model information and recognizing the source language as a word string. The result of speech recognition by the speech recognition unit 2 is output to the translation unit 5 as the word string candidate.

At this time, the voice recognition unit 2 generates a word string candidate by adding a voice recognition score to the generated word string.
The speech recognition unit 2 may express the word string candidate as a speech recognition result in a graph structure as shown in FIG.

The speech recognition score added to the word string candidate by the speech recognition unit 2 includes a score for each word when speech recognition processing is performed using acoustic model information when a speech signal is detected, and a recognition language. Each word is connected by the model information to generate a word string candidate.

The speech recognition unit 2 generates a score when using the acoustic model information using, for example, an HMM, and generates the above-mentioned speech recognition score when using the recognition language model information, such as a word chain probability. It is generated by using commonly used technology. Also, this voice recognition unit 2
Generates a speech recognition score by performing processing using a probabilistic sentence speech recognition algorithm described in, for example, a document “Seiichi Nakagawa,“ Speech Recognition by Probabilistic Model ”, The Institute of Electronics, Information and Communication Engineers, Corona”.

The translation unit 5 linguistically analyzes the word sequence candidates from the speech recognition unit 2 using the translation language rule information stored in the translation language rule storage unit 6 for the word sequence candidates from the speech recognition unit 2. And a translation process for performing a target language generation process for generating a target language. Here, the translation unit 5 performs the translation processing by executing, for example, a translation processing algorithm stored in an internal memory. Thus, the translation unit 5 outputs the speech of the source language input to the speech recognition unit 2 by performing character data translated into the target language or speech synthesis processing. The translation unit 5 generates a translation score for each candidate of the translation processing result, and outputs the translation score to the composite score calculation unit 7.

The translation unit 5 performs the above-mentioned translation using various costs, such as the probability of a rule for language analysis, the probability of a grammar for generation, and the word chain probability of a surface of a generated word in a target language. Generate a score. This translation unit 5
The process of generating the translation score by using not only the process using the above-described probability model but also the content may be adjusted by a user, for example.

The translation language rule storage unit 6 causes the translation unit 5 to output the stored translation language rule information by reading the contents of the translation language rule. This translation language rule information is information for recognizing the part of speech of each word and generating a sentence and a phrase formed by connecting a plurality of words. The translation language rule information includes, for example, information indicating a rule in a form of “generating a noun phrase by connecting an adjective and a noun”, and word dictionary information indicating semantic information of each word. Further, as shown in FIG. 5, this translation language rule information describes, for example, "sentence → noun phrase 'is'" and indicates that the "sentence" is composed of "noun phrase" and "is", and "noun phrase" The phrase "noun phrase" is described as "noun phrase", and indicates that "noun phrase" is composed of "noun phrase", "ha", "noun phrase", and the like. In other words, the translation language rule information describes the connection relation of phrases and parts of speech for forming sentences, and the connection relation of phrases or parts of speech for forming phrases.

Using the translation language rule information shown in FIG. 5, the translation unit 5 converts the word string candidate represented by the graph structure shown in FIG. 4 into a syntax tree as shown in FIG. The translation process is performed by creating it.

The synthetic score calculation unit 7 performs a calculation using the acoustic score from the speech recognition unit 2 and the translation score from the translation unit 5, thereby translating the input speech and outputting the translation from the translation unit 5. Generate a composite score for the result. At this time, when the acoustic score and the translation score are strictly calculated according to the above-mentioned probability model, the synthetic score calculating unit 7 performs the calculation of taking the sum of the product or the logarithm of the acoustic score and the translation score, thereby obtaining the synthetic score. Is performed. Then, the composite score calculation unit 7 adds the generated composite score to each of the plurality of translation processing results and outputs the result.

Further, the composite score calculation section 7 performs a process of determining whether or not to output a translation processing result according to the value of the composite score. For example, when the combined score is determined to be equal to or more than the predetermined value, the combined score calculation unit 7 does not perform the process of outputting the translation processing result, and when the combined score is determined to be equal to or less than the predetermined value, outputs the translated processing result to which the combined score is added. Perform processing. Further, the composite score calculation unit 7 may perform a process of sequentially outputting the translation processing results having higher composite scores in accordance with the generated composite score.

Thus, the synthetic score calculation section 7 generates a synthetic score according to the probability model, and performs processing in the voice recognition section 2 and the translation section 5 from generation of a source language voice to a target language character string. The translation processing results that have been passed are output in descending order of the composite score generated based on the adopted probability model.

The speech recognition / translation apparatus 1 configured as described above
Has a synthesized score calculation unit 7 that generates a synthesized score using the speech recognition score in the speech recognition process as well as the translation score in the translation process as described above. Can be output after adding a score in consideration of voice recognition processing. Therefore, according to the speech recognition and translation device 1, since the synthesis score considering the speech recognition process is added and output, the accuracy of the translation process result for the input speech can be improved.

The composite score calculation unit 7 generates a composite score by weighting the speech recognition score in the speech recognition process of the speech recognition unit 2 and the translation score in the translation process of the translation unit 5. Is also good. That is, when the speech recognition score is X, the translation score is Y, and the weighting factors for each are A and B, the synthetic score calculation unit 7 sets the evaluation function as AX + BY. Then, the synthesis score calculation unit 7 sets the weighting coefficients A and B and applies the above evaluation function to generate a synthesis score reflecting the speech recognition score and the translation score.

Thus, when the synthetic score is not a strict probability model, or when the accuracy of the speech recognition process and the translation process are not the same, the synthetic score calculation unit 7 gives priority to the score in the speech recognition process or the translation process. A process for generating a composite score is performed to improve the processing accuracy.

It should be noted that the evaluation function is not limited to the above example, but may be another expression. That is, the synthetic score calculation unit 7 applies the evaluation function that appropriately reflects the voice recognition score in the voice recognition process to the synthetic score for the translation process result, thereby obtaining the translation process result including the plurality of word strings. , It is possible to increase the possibility of selecting a correct translation processing result.

The present invention can also be applied to a speech recognition and translation device 10 having, for example, the configuration shown in FIG. The same parts as those of the speech recognition and translation device 1 shown in FIG. 1 are denoted by the same reference numerals, and detailed description thereof will be omitted.

The speech recognition translator 10 is generated by calculating a synthetic score using the linguistic analysis generation unit 11 to which the speech recognition result is input from the speech recognition unit 2 and the linguistic analysis result from the linguistic analysis generation unit 11. And a translating unit 13 comprising a synthetic score calculating unit 12 for performing the translation.

The language analysis generation unit 11 is connected to the translation language rule storage unit 6, reads the translation language rule information from the translation language rule storage unit 6, and applies the translation processing algorithm to the language analysis generation unit 11. The speech recognition result is subjected to a translation process including a language analysis process for creating a syntax tree as shown in FIG. 6 and a target language generation process. Then, the linguistic analysis generation unit 11 outputs the translation processing result expressed in the generated target language via the synthesis score calculation unit 12.

Further, the linguistic analysis generation unit 11 outputs to the synthesis score calculation unit 12 intermediate processing information indicating an intermediate result in the linguistic analysis processing by executing the above-described translation processing algorithm. In addition, the language analysis generation unit 11
Generates an intermediate processing score for the intermediate processing information, adds the generated intermediate processing information to the intermediate processing information, and outputs it to the composite score calculation unit 12.

The speech recognition score is input from the speech recognition unit 2 to the synthesis score calculation unit 12 together with the intermediate processing information and the translation processing result from the language analysis generation unit 11. When the translation processing result is input from the language analysis generation unit 11, the synthesis score calculation unit 12 generates a synthesis score using the translation score and the speech recognition score added to the translation processing result. Then, similarly to the translation unit 5 described above, the synthesis score calculation unit 12 performs a process of generating a synthesis score using the translation score and the speech recognition score, and outputs a translation processing result according to the value of the synthesis score. A process is performed to determine whether or not this is the case.

When the intermediate processing information is input, the synthetic score calculating unit 12 predicts a synthetic score to be added to the translation processing result using the intermediate processing score and the speech recognition score added to the intermediate processing information. And performs a score prediction process of generating a prediction score. Then, the composite score calculation unit 12 outputs the prediction score obtained by performing the score prediction process to the language analysis generation unit 11.

The language analysis generation unit 1 of the translation unit 13
When the prediction score is input from the synthetic score calculation unit 12, 1 determines whether to stop the translation process according to the prediction score.

For example, when a translation process is performed on the speech recognition result from the speech recognition unit 2, the prediction score for the in-process information in which the application of the syntax rule that is very unlikely to occur in the language analysis is continuous is extremely low. Become. That is, based on such intermediate processing information, the language analysis generation unit 1
Even if the translation processing is performed in step 1, it is unlikely that a correct translation processing result is obtained.

At this time, according to the speech recognition and translation device 10 including the above-described translation unit 13, the linguistic analysis generation unit 11 generates the prediction score by the synthesis score calculation unit 12 using the intermediate processing score and the speech recognition score. Thus, the language analysis generation unit 11 can perform a process of referring to the prediction score to stop the translation process by the translation process algorithm or a language analysis process for another word string candidate. Also,
According to the speech recognition / translation device 10, the synthetic score calculation unit 12 generates a prediction score using not only the midway processing score but also the speech recognition score, and the language analysis generation unit 11 performs a process of stopping the translation process. In addition, it is possible to more accurately determine whether to stop the translation process in the linguistic analysis generation unit 11.

More specifically, the linguistic analysis generation unit 11 outputs, for example, when a voice saying “My name is Yamada” is input to the voice recognition unit 2 and is translated and output as “My name is yamada”. The intermediate processing result after performing the language analysis processing on the noun phrase composed of “my name” in FIG. 6 and the intermediate processing score for the intermediate processing result are calculated by the translation processing algorithm. Then, the synthetic score calculation unit 12 performs a process of generating a prediction score using the intermediate processing result, the intermediate processing score, and the speech recognition score of the word string candidate for the noun phrase, and outputs the predicted score to the language analysis generation unit 11. Thereby, the language analysis generation unit 11
Determines that the process is terminated if the predicted score from the combined score calculation unit 12 is equal to or less than a certain value, and determines that the process is to be continued if the predicted score is equal to or more than a certain value.

Further, in the above-described speech recognition and translation apparatus 10, the language analysis generation unit 11 is not limited to an example of the processing for outputting the intermediate processing result and the intermediate processing score in the language analysis processing to the synthesis score calculation unit 12, and may include, for example, a target language. May be generated and the intermediate processing score for the intermediate processing result may be generated. For example, when the appearance probability (score) of the used translation language rule information is very low, and when the appearance probability of the word string of the target language appearing on the surface is low, and the speech recognition processing of the source language corresponding to the part concerned, When the voice recognition score is low, the translation process for the word string candidate may be stopped.

In the speech recognition and translation apparatus 10, a word in the target language (word string) and a word in the source language (word string) are obtained by translating one word into a plurality of words or changing the word order greatly. In some cases, a clear response cannot be taken. However, there are many cases where correspondence can be relatively easily made between nouns and the like, and it is considered that generating a synthetic score using the speech recognition score and the recognition score is effective for improving the processing accuracy of the translation processing result.

In the speech recognition / translation apparatus 10, the processing for stopping the translation processing in the language analysis generation unit 11 is as follows.
Of course, the processing may be performed using only the intermediate processing score.

Next, an example in which the translation unit 13 of the speech recognition translation apparatus 10 described with reference to FIG. 7 performs a translation process according to a translation algorithm, thereby stopping the translation process according to the intermediate processing result. Will be described with reference to the flowchart shown in FIG. The language analysis processing unit 11 performs the translation process by performing the language analysis process and the target language generation process. In the following description of the flowchart of FIG. 8, a certain word input from the speech recognition unit 2 is used. An example in which the translation process is executed until the completion of the translation process for the column candidates or until it is determined that the process is stopped based on the prediction score will be described.

According to FIG. 8, first, in step S1, the language analysis processing section 11 outputs the translation processing result to the outside by performing the above-described language analysis processing and target language generation processing by the translation section 13. It is determined whether the translation process has been completed. Then, when the language analysis processing unit 11 determines that the translation process has been completed, the process ends, and when it determines that the translation process has been completed, the process proceeds to step S2.

In the next step S2, the linguistic analysis generation unit 11 performs linguistic analysis processing for one step. Here, the linguistic analysis processing for one step refers to, for example, processing of applying the translation linguistic rule information used when performing linguistic analysis to the word string candidate from the speech recognition unit 2 only once.

In the next step S3, the linguistic analysis generation unit 11 determines whether or not to perform the above-mentioned processing continuation check on the intermediate processing result after performing the above-mentioned processing in step S2 in the above-mentioned translation processing algorithm. That is, the linguistic analysis generation unit 11 determines whether or not to perform the determination to stop the processing according to the intermediate processing score for the intermediate processing result. Then, the language analysis generation unit 1
1 proceeds to step S4 when performing the processing continuation check, and proceeds to step S4 when not performing the processing continuation check.
Return to 1. That is, the linguistic analysis generation unit 11 performs the processing in steps S1 to S1 regardless of the intermediate result until the processing stage in which it is determined in step S3 to perform the processing continuation check.
By repeating step S2, the translation process is continuously performed according to the above translation algorithm.

In the next step S4, the linguistic analysis and generation unit 11 converts the translated language rule information used for creating the syntax tree, the score of the translated language rule information, and the word string candidate to be translated. The intermediate processing score calculated from the acoustic score is generated, and the intermediate processing result in the linguistic analysis processing is output to the synthetic score calculation unit 12 together with the intermediate processing score. Then, the composite score calculation unit 12
Calculates a prediction score using the intermediate processing score from the language analysis generation unit 11 and the acoustic score to be translated, and outputs the prediction score to the language analysis generation unit 11.

In the next step S5, when the predicted score is input from the synthesized score calculation unit 12, the language analysis generation unit 11 determines whether or not to continue the translation process according to the predicted score. Then, when the linguistic analysis generation unit 11 determines that the translation process is not to be continued, the process ends, and when it determines that the translation process is to be continued, the process returns to step S1. That is, the language analysis generation unit 11
Means that by repeating steps S1 to S5, the translation process is executed until the process is stopped based on the prediction score.

The speech recognition and translation device 10 provided with the translation unit 13 for performing such a process, when performing the translation process,
Since the combined score calculation unit 12 generates a prediction score and performs a processing continuation check using the intermediate processing score and the like for the intermediate processing result at each predetermined step, the translation processing result is compared with the intermediate processing result during the language analysis processing. The processing can be terminated for the word string candidate having a low combined score. Therefore, according to the speech recognition / translation apparatus 10, it is possible to reduce the processing for word string candidates that do not need to be translated, and to reduce the processing amount of the translation processing for the input speech, thereby achieving high-speed translation processing. It can be performed.

The processing for one step in step S2 described above differs depending on the translation processing algorithm. The translation processing algorithm in FIG. 8 is a processing step arbitrarily determined during the language analysis processing in the translation processing. The processing unit for performing the processing continuation check after passing through is one step.

[0054]

As described above in detail, according to the apparatus and method for speech recognition and translation according to the present invention, the synthesis which indicates the processing accuracy of the speech recognition and translation using the speech recognition score and the translation score. Since the score is generated and the translation result to be output is evaluated using the composite score, the accuracy of evaluation of the translation result can be improved. Therefore, according to the apparatus and method for speech recognition and translation according to the present invention, it is possible to improve the processing accuracy by using the respective scores in the speech recognition processing and the translation processing. According to the speech recognition translation apparatus and method, a score for a translation result is generated using a speech recognition score for speech recognition processing. For example, a translation processing score is calculated during the translation processing to generate a synthetic score. However, the translation process can be stopped halfway according to the synthesis score, unnecessary processes in the translation process can be reduced, and the speed of the translation process can be increased.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a speech recognition and translation device to which the present invention has been applied.

FIG. 2 is a diagram for describing acoustic model information and word dictionary information.

FIG. 3 is a diagram for describing recognition language model information.

FIG. 4 is a diagram for describing an output result of a graph structure output from a speech recognition unit.

FIG. 5 is a diagram for explaining translation language rule information.

FIG. 6 is a diagram illustrating that a translation unit performs a translation process by creating a syntax tree.

FIG. 7 is a block diagram showing a configuration of another speech recognition and translation apparatus to which the present invention has been applied.

FIG. 8 is a flowchart illustrating a process in which a translation unit performs a translation process according to a translation algorithm, and performs a process of stopping the translation process in accordance with a result of the intermediate process.

[Explanation of symbols]

1 speech recognition and translation device, 2 speech recognition unit, 5 translation unit,
7 Composite score calculation unit, 11 Language analysis generation unit, 12
Composite score calculation unit, 13 translation unit

Claims (8)

[Claims] 1. Speech recognition by applying, to an inputted speech, an acoustic model referred to for recognizing a waveform shape of a speech signal as speech and a recognition language model indicating a connection relation of each word constituting a sentence. Voice recognition means for performing processing; voice recognition score generation means for generating a voice recognition score indicating processing accuracy of the voice recognition processing with respect to the voice recognition result from the voice recognition means; voice recognition result from the voice recognition means A translation unit that performs a translation process for translating into a language different from the voice input using, a translation score generation unit that generates a translation score indicating the processing accuracy of the translation process for a translation result from the translation unit, Using the voice recognition score from the voice recognition score generating means and the translation score from the translation score generating means, the voice recognition processing and translation A synthetic score generating means for generating a synthetic score indicating processing accuracy of the processing, and an evaluating means for evaluating a translation result output from the translating means by using the synthetic score from the synthetic score generating means. Speech recognition translator. 2. The translation means according to claim 1, wherein said translation means performs a language analysis on a speech recognition result from the speech recognition means, and a language constituting a language different from the speech input by the language analysis result from said language analysis means. Generating a translation score by generating a score for the linguistic analysis result from the linguistic analysis unit and a score for the language configured by the linguistic generation unit. The speech recognition and translation device according to claim 1, wherein: 3. An intermediate processing score generating means for generating an intermediate processing score for the intermediate processing result of the translation processing in the translation means, and whether or not to continue the translation processing based on the intermediate processing score from the intermediate score generating means. 2. The speech recognition and translation device according to claim 1, further comprising: an intermediate processing score determination unit that determines the score. 4. Using the intermediate processing score from the intermediate processing score generating means and a speech recognition score for a speech recognition result to be translated by the translating means, a translation score for the translation result is calculated. A prediction score generation unit that generates a prediction score obtained by prediction; and a prediction score determination unit that determines whether to continue the translation process based on the prediction score from the prediction score generation unit. The speech recognition translator according to claim 3. 5. A speech recognition process for applying, to an inputted speech, an acoustic model referred to for recognizing a waveform shape of a speech signal as a speech and a recognition language model indicating a connection relationship between words constituting a sentence. A voice recognition score generation process for generating a voice recognition score indicating the processing accuracy of the voice recognition process for the voice recognition result obtained by the voice recognition process; and using the voice recognition result to a language different from the input voice. A translation process for translating; a translation score generation process for generating a translation score indicating a processing accuracy of the translation process for a translation result by the translation process; and a speech recognition process using the speech recognition score and the translation score. And a synthetic score generating process for generating a synthetic score indicating the processing accuracy of the translation process, and evaluating the output translation result using the synthetic score. Speech recognition translation method and having an evaluation process for. 6. The translation process includes a language analysis process for performing a language analysis on the speech recognition result, and a language generation process for configuring a language different from the speech input based on the language analysis result. 6. The speech recognition translation method according to claim 5, wherein in the generation process, a translation score is generated by generating a score for the language analysis result and a score for the language configured by the language generation process. 7. An intermediate process score generating process for generating an intermediate process score for the intermediate process result of the translation process, and an intermediate process score determining process for determining whether to continue the translation process based on the intermediate process score. 6. The speech recognition translation method according to claim 5, comprising: 8. A prediction score obtained by predicting a translation score for a translation result using the intermediate processing score and a speech recognition score for a speech recognition result to be subjected to the translation process. The speech recognition translation method according to claim 7, further comprising: a prediction score generation process; and a prediction score determination process for determining whether to continue the translation process based on the prediction score.